Global Journal of Computer Science and Technology, C: Software & Data Engineering, Volume 22 Issue 2

Fake News Detection: Covid-19 Perspective Global Journal of Computer Science and Technology Volume XXII Issue II Version I 9 Year 2022 ( ) C © 2022 Global Journals Technique Algorithms T/F Support Precision Recall F1-Score Accuracy TF-IDF Multinomial na¨ıve bias Set-1 0 53 0.93 0.81 0.87 0.89 1 63 0.86 0.95 0.9 Set-2 0 63 0.97 0.46 0.62 0.73 1 69 0.67 0.99 0.8 Set-3 0 61 1 0.23 0.37 0.72 1 104 0.69 1 0.82 Set-4 0 53 1 0.08 0.14 0.81 1 211 0.81 1 0.9 Passive Aggressive Set-1 0 53 0.87 0.87 0.87 0.88 1 63 0.89 0.89 0.89 Set-2 0 63 0.93 0.86 0.89 0.9 1 69 0.88 0.94 0.91 Set-3 0 61 0.9 0.77 0.83 0.88 1 104 0.88 0.95 0.91 Set-4 0 53 0.92 0.64 0.76 0.92 1 211 0.92 0.99 0.95 Support Vector Machine Set-1 0 53 0.84 0.87 0.85 0.86 1 63 0.89 0.86 0.87 Set-2 0 63 0.93 0.83 0.87 0.89 1 69 0.86 0.94 0.9 Set-3 0 61 0.96 0.74 0.83 0.89 1 104 0.86 0.98 0.92 Set-4 0 53 0.89 0.62 0.73 0.91 1 211 0.91 0.98 0.95 Logistic Regression Set-1 0 53 0.84 0.87 0.85 0.86 1 63 0.89 0.86 0.87 Set-2 0 63 0.95 0.83 0.88 0.89 1 69 0.86 0.96 0.90 Set-3 0 61 0.90 0.77 0.83 0.88 1 104 0.88 0.95 0.91 Set-4 0 53 0.91 0.6 0.73 0.91 1 211 0.91 0.99 0.95 Table 2: Fake News Detection Result Using TF-IDF with logistic regression using count vectorizer also shown best performance on accuracy as we can see in Table 3 but the poor result on other three-parameter. The next highest accuracy of 89% is given by set 2 and set 3. Where set 2 given higher precision (90%), recall (86%), and f1-score (88%) at detecting fake news. In last we have our final result of support vector machine using TF-IDF given in Table 2. In here also set 4 shown the highest accuracy (91%). But poor performance on the other three-parameter. Set 2 and set 3 shown better performance as we can see in Table 2. Set 2 shown a precision of 93%, recall of 83%, f1-score of 87%, and accuracy of 89% which is combined best the other three sets. In Table 3, support vector machine using count vectorizer, as usual like other three algorithm support vector machine has shown highest accuracy of 90% in set 4 but the same poor result on other three-parameter. Set 3 shown the second highest accuracy of 88% and 84% on recall, precision, and f1-score. Set 2 gives an accuracy of 87%, the third-highest among four algorithms but having the higher value of precision (86%), recall (87%), and f1-score (87%) the other three algorithms. Using set 4, almost all algorithms have given the using TF-IDF. In all algorithms either using TF-IDF or count vectorizer method set 4 have given better accuracy but poor performance on other three- parameter. Set 1 and set 2 given decent and stable values. But set 2 was better in all algorithms either using TF-IDF or the count vectorizer method. So, from our observation, we can say that an excessive amount of one type of data can be misleading while detecting fake news. For our further analysis, we have taken set 2, as our final comparison shown in Fig. 10. Using the TF-IDF method Multinomial how a satisfying result. But other three algorithms showed better outcomes Fig. 10b. On the other hand, using the count vectorizer method all of the four algorithms shown better outcomes as we can see in Fig. 10a VI. C onclusion Fake news is one of the alarming issues of the digital era and fake news detection can halt this issue. Through our work, we want to contribute to solve this issue by highest accuracy except Multinomial bayes Bayes does not s Naïve Naïve